Cloud Hosting

With less than a week to go, the excitement and anticipation are building up for industrys largest cloud computing conference - AWS re:Invent.

Las Vegas

As an analyst, I have been attempting to predict the announcements from re:Invent (2018, 2017) with decent accuracy. But with each passing year, its becoming increasingly tough to predict the year-end news from Vegas. Amazon is venturing into new areas that are least expected by the analysts, customers, and its competitors.

AWS Ground Station is an example of how creative the teams at Amazon can get in conceiving new products and services. Announced at last years re:Invent, AWS Ground Station is a managed service to control satellite communications and process the downlink data. This service took everyone by surprise. I wonder if AWS is planning to launch a managed air traffic controller (ATC) service for airports all over the world to manage air traffic.

Along with a few predictions, I want to share my wishlist for re:Invent 2019. This is based on my personal experience of using AWS combined with the feedback from my customers.

Here is what I expect to see from AWS re:Invent 2019:

Streamlined Developer Experience

AWS now has multiple compute services in the form of EC2 (IaaS), Beanstalk (PaaS), Lambda (FaaS) and Container Services offered through ECS, Fargate and EKS (CaaS). Modern applications rely on more than one compute model for execution. For example, its a common practice to deploy containerized applications on EKS while running event-driven code in Lambda. Legacy applications continue to target Amazon EC2 instances.

Within the container services, ECS, EKS and Fargate use different approaches and patterns of deployment. Fargate is a node-less environment while EKS is a managed Kubernetes service.

Amazon customers need a better experience of deploying and managing applications on AWS. A unified framework that abstracts underlying compute services will simplify packaging and deploying applications. This framework may be an extension of CloudFormation, Kubernetes YAML, Cloud Developer Kit (CDK) and Serverless Application Model (SAM).

Its time for Amazon to simplify consuming the compute services of AWS.

Kubernetes-based Hybrid Cloud

Going by the industry trend, I expect AWS to leverage Kubernetes for hybrid cloud. There may be a new control plane to manage EKS clusters running in the cloud along with the non-EKS clusters running on-premises. Customers will be able to deploy applications and apply configuration settings to managed EKS clusters, EKS clusters running on AWS Outposts and even unmanaged Kubernetes clusters running within the data center. I doubt if AWS will go the Google Anthos and Azure Arc way to support multi-cloud environments but integrating Kubernetes clusters with a single control plane will help enterprise customers. This will also enable VMware environments running Project Pacific and PKS deployments seamlessly integrate with Amazon EKS.

App Model for Kubernetes

With the new found love for open source, Amazon may announce an OSS project targeting Kubernetes much on the lines of Knative and Rudr. Google is investing in Knative as the platform for Kubernetes while Microsoft is building Rudr as the application layer of Kubernetes.

New Managed Services for VMware and Outposts

Additional managed services such as DynamoDB and Lambda may become available on VMware Cloud on AWS and Outposts.

Finally, there will be new EC2 instances optimized for deep learning training and inference. New instance types based on ARM processors may also be announced this year.

AutoML for Custom Machine Learning Models

Though AWS has a solid ML PaaS in the form of SageMaker and a set of AI services, it lacks AutoML - a capability that simplifies training deep learning models on custom datasets. Amazon Rekognition Custom Labels is a step towards AutoML for vision. I expect AWS to add AutoML support for video, text classification, translation, and even tabular data.

Custom Processor for ML Training & Inference

Currently, AWS relies on NVIDIA GPUs for training deep learning models. For inference, it uses both NVIDIA GPUs for Elastic Inference and a purpose-built machine learning chip for AWS Inferentia.

With its investments in Project Nitro, Amazon is expected to build an Application Specific Integrated Circuit (ASIC) optimized for training and inference. The chips based on the ASIC will power a new family of EC2 instances and also a subset of AWS Outposts.

AWS will be able to offer a cheaper training and inferencing service compared to GPU-based environments. AWS may fork Apache MXNet, TensorFlow, and PyTorch to build optimized versions of frameworks targeting the ASIC.

Hosted ML Project Management

To support managing, tracking, and sharing ML projects, SageMaker may get a hosted ML management tool. This will be modeled on the lines of Databricks MLflow and Azure ML Services. Even those ML experiments running outside of SageMaker may consume the service through an API and SDK.

Autonomous Systems based on Reinforcement Learning

Amazon is one of the first public cloud providers to bet on reinforcement learning. It launched highly successful DeepRacer device and a racing competition last year.

This year, AWS may get a managed reinforcement learning platform to build autonomous systems. The new deep reinforcement learning service will enable domain experts in the fields of medical, automobile, energy, and manufacturing to build sophisticated models.

Amazon Comprehend for Legal and Finance

Having launched Comprehend Medical, I expect AWS to extend Comprehend to legal and finance domains. This service will enable customers to extract domain-specific terminology and nomenclature from unstructured data. Law firms, accounting professionals, and stock broking companies will benefit from this service.

Amazon Transcribe with Custom Speaker Identification

Amazon Transcribe, the speech-to-text service may be able to identify and recognize speakers based on custom datasets with voice clips. This enhancement will enable developers to build user profiles and personalisation based on speech and voice recognition.

Cheaper Jupyter Notebooks on GPU-backed Spot Instances

The only way customers launch Jupyter Notebooks on AWS is through Deep Learning AMI or Amazon SageMaker. AWS may make it easy and cheap to launch Jupyter Notebooks by hosting them on Spot Instances backed by GPUs. Though it may not be free, the service will be modeled around Google Colab.

ONNX Support for SageMaker Neo

Amazon is one of the founding members of the Open Neural Network Exchange (ONNX) initiative. ONNX aims to bring interoperability to deep learning frameworks by enabling developers to import and export models from one framework to the other. Amazon SageMaker Neo is a runtime based on Apache TVM to run machine learning models across the cloud and edge. AWS may finally announce the support for ONNX for SageMaker Neo. This makes it possible to build deployment pipelines that target a variety of edge environments including mobile phones and desktops.

AIOps for CloudWatch

Amazon is enhancing CloudWatch to support modern observability patterns. With AIOps, CloudWatch may be able to detect anomalies based on the logs that are ingested. It can even perform root cause analysis (RCA) when the workload experiences disruption.

GPU-enabled Data Pipelines

Currently, none of the data processing and ETL services on AWS support parallelized processing taking advantage of GPUs. Much on the lines of NVIDIA Rapids, AWS may enable GPU support for data processing pipelines. This may get extended to query processing engines powering Amazon RDS, Amazon Aurora, and Redshift.

SaaS Service for IoT

Amazon has a wide range of services related to IoT and Edge. However, it lacks a SaaS-based IoT offering to easily connect and manage devices. There may be a new IoT service designed on the lines of Azure IoT Central.

I am looking forward to re:Invent 2019! Stay tuned as I continue to bring the analysis and commentary from the biggest cloud event of the year.

Read this article:
AWS re:Invent 2019 - Predictions And A Wishlist - Forbes